Thursday, May 03, 2007

The Future of Image Search Belongs to Social Search

from Thomas Hawk's Digital Connection by Well, image search is one of the hardest types of search (audio and video aren't so easy either). With text search, Google, Yahoo and Microsoft all have their proprietary algorithms where they look for text on a page, see who links to a page, etc. etc. words are in contrast to images much easier to figure out. If Mike Arrington is mentioned 40 times in a post by a highly ranked internet site, then the article probably has some authority to be placed in the results for an article about Mike Arrington. But photos of Mike Arrington are a different matter. It is very difficult for image search engines to get at what's inside a photo and how good a quality photo it is. Accordingly, image search engines that rely solely on algorithms without any human filtering fall flat compared to results that are filtered through social networks.

MaryDonna [I'm CEO of Zooomr, we are building both a social based image search system as well as a stock photography platform] Live Image Search advances | Larry Larsen | Channel 10 Larry Larsen over at Channel 10 blogs today about some recent enhancements that Microsoft has made to their image search technology and suggests that they have "greatly enhanced relevance," and as such deserve a "day off." Unfortunately, I'm going to have to disagree. While I like the fact that Microsoft claims an increase in speed on how fast their images load, the relevancy of their results still pale significantly in comparison to what can be done with social search. This is not the first time that I've blogged about this and it won't be the last. The future of image search very much belongs to social search. What do I mean by this? Well, image search is one of the hardest types of search (audio and video aren't so easy either). With text search, Google, Yahoo and Microsoft all have their proprietary algorithms where they look for text on a page, see who links to a page, etc. etc. words are in contrast to images much easier to figure out. If Mike Arrington is mentioned 40 times in a post by a highly ranked internet site, then the article probably has some authority to be placed in the results for an article about Mike Arrington. But photos of Mike Arrington are a different matter. It is very difficult for image search engines to get at what's inside a photo and how good a quality photo it is. Accordingly, image search engines that rely solely on algorithms without any human filtering fall flat compared to results that are filtered through social networks. To see what I mean lets look at some examples: Mike Arrington, Flickr Mike Arrington, "new and improved" live.com Mike Arrington, Yahoo Image Search Mike Arrington, Google Image Search Mike Arrington, Ask.com Summer, Flickr Summer, "new and improved" live.com Summer, Yahoo Image Search Summer, Google Image Search Summer, Ask.com (I particularly appreciated the relevance of that third row result on Flickr). Brunette, Flickr Brunette, "new and improved" live.com Brunette, Yahoo Image Search Brunette, Google Image Search Brunette, Ask.com (ok, so which would you like to date the most, isn't the difference between Ask and Yahoo dramatic?) Africa, Flickr Africa, "new and improved" live.com Africa, Yahoo Image Search Africa, Google Image Search Africa, Ask.com As you can see from the examples above, the higher quality, better caliber images generally come from Flickr. Flickr's results are screened through their social network. The users validate which photos are best by their social activity around the photographs. Users also tag photos to better identify what's inside the photo. Yahoo's image search is largely the worst. This is the future of image search. It is also, by the way, the future of the $2.5 billion stock photography market. Comparable searches between Getty Images, Corbis and Flickr would produce comparable results. This is why we are working on building the best stock photography search engine in the world on Zooomr right now. It will certainly have application for broader more generic public image search, but it most certainly will be the future of the stock photography business as well.